The United States government has managed to do something genuinely impressive: it blacklisted Anthropic and became its biggest cheerleader at the same time.

Here is the timeline. On April 8, a federal appeals court declined to temporarily block the Pentagon's supply-chain risk designation against Anthropic, leaving the blacklist firmly in place. The company remains excluded from all new Department of Defense contracts and systems. That designation came after Anthropic refused to let the military use Claude without guardrails against mass surveillance and autonomous weapons.

On the same week, CNBC reported that Vice President JD Vance and Treasury Secretary Scott Bessent personally encouraged major banks to test Anthropic's Mythos model for cybersecurity defense. The model that the Pentagon says is too dangerous for contractors to touch is apparently exactly what the Treasury Department wants protecting America's financial system.

And on Monday, Reuters reported that Dario Amodei, Anthropic's co-founder and CEO, confirmed the company is in active discussions with the Trump administration about Mythos. Despite. The. Blacklist.

The legal technicality that makes this possible is narrower than it appears. The Pentagon's supply-chain risk designation only blocks Anthropic from DoD contracts. It does not prevent other federal agencies from working with the company. So the Treasury can call Anthropic, the Fed can test Mythos, banks can pilot the model at government encouragement, and none of that violates the blacklist. The Pentagon and the Treasury are operating in parallel universes on the same AI company.

This is not just bureaucratic absurdity. It reveals something fundamental about where AI power actually lives right now. Anthropic built a model so capable at finding software vulnerabilities that it triggered emergency meetings with the Treasury Secretary and the Fed Chair. It forced the White House to call every major AI CEO to a secret briefing. Banks, governments, and critical infrastructure operators all need it. And the one branch of government that said no is the Pentagon, which blacklisted Anthropic not because the model is bad but because Anthropic would not let the military use it without restrictions.

Put differently: the government punished Anthropic for having safety standards, then came running back because those same safety standards produced a model good enough to defend the country's financial infrastructure.

The Guardian published a blistering analysis this week calling Anthropic's Mythos rollout a potential bait and switch, suggesting the company is using safety as a PR tool before prioritizing profits. Security researcher Hoda Khlaaf pointed out that Anthropic's decision to limit public access prevents independent experts from evaluating the company's claims. That is a legitimate criticism. Responsible AI claims that cannot be independently verified are just marketing.

But even the skeptics are engaging with the model. Apple, Google, Microsoft, and JPMorgan all have access through Project Glasswing. Canada called an emergency meeting about it. The Bank of England summoned bank executives. Britain, Australia, and Japan have all opened discussions.

Anthropic has turned a Pentagon blacklist into the greatest product launch in AI history. The government told everyone not to use it, and everyone took that as a recommendation.

Whether this is genuine safety leadership or the most sophisticated marketing play in tech, the result is the same: the company that got kicked out of the Pentagon is now in conversations with every major government on Earth. And the legal case is still very much alive. If Anthropic wins on appeal, it gets the Pentagon back too. If it loses, it has already proven it does not need the Pentagon to dominate.